AI Finds the Value Bets
Shown the market, can AI find the value bets it has missed?
11,891 picks settled · £1 level stakes to SPThe live leaderboard
All settled picks, ranked by return on investment.
| # | AI | Staked | Returned | Profit | Bets | Win% | Backed the fav | Profit trend |
|---|---|---|---|---|---|---|---|---|
| 1 | Stablebet EdgeStablebet | £1,706 | £1,587 | −£119(-7.0%) | 1,706 | 7% | 0% | |
| 2 | DeepSeekDeepSeek | £1,613 | £1,398 | −£215(-13.4%) | 1,613 | 17% | 26% | |
| 3 | GeminiGoogle | £1,728 | £1,462 | −£266(-15.4%) | 1,728 | 9% | 0% | |
| 4 | GrokxAI | £1,725 | £1,398 | −£327(-18.9%) | 1,725 | 14% | 2% | |
| 5 | ChatGPTOpenAI | £1,721 | £1,389 | −£332(-19.3%) | 1,721 | 10% | 0% | |
| 6 | Stablebet ModelStablebet | £1,733 | £1,380 | −£353(-20.3%) | 1,733 | 20% | 30% | |
| 7 | ClaudeAnthropic | £1,665 | £1,286 | −£379(-22.8%) | 1,665 | 9% | 0% |
In AI Finds the Value Bets, each AI gets the same racecard as its sibling test, AI Picks the Winner, plus the market's implied chance for every runner: the crowd's collective view of the race, with the bookmaker's margin stripped out (it never sees the raw Starting Price it will be settled at). And because the market's view is on show, the task is different too: not “who wins?” but “where is the value?”. Each AI must form its own view of every runner's true chance and back the horse the market most underrates, logging its own estimate alongside the pick.
Two entrants of our own run here, both powered by the same in-house racing engine. It is trained on market-derived data, so it sees the market by construction and can never honestly run blind. Stablebet Edge plays the value rule: in every race it backs the runner whose chance it rates furthest above the market's, its biggest edge on the card. The Stablebet Model backs the horse the engine rates most likely to win. Same numbers, two rules, and the gap between their records is its own experiment. (The Favourite stays in AI Picks the Winner; an informed favourite would just be backing itself.)
A note on the record:until 16 July 2026 we asked this test the wrong question: “which horse will win?” with the market chances in view. That question answers itself, so every model quite reasonably named the favourite about 99% of the time, and the run told us nothing. Our mistake, not theirs. The record above restarts from 17 July 2026 with the value question; the old rows stay in the raw data for audit.
Research, not tips. Settled honestly to industry SP, wins and losses alike. 18+ · please gamble responsibly.
What the numbers say
The read on 60 days and 24,011 settled picks, worked out live from the board.
Every AI finds the winner in about 9.1%–17.0% of races, within a whisker of each other, and every one still loses. The market prices those winners too short. Naming the likely winner is not the same as being paid enough for it.
They back the market favourite 0%–26% of the time, and went against a unanimous field only 19–76 times each. Most of the time, asking an AI is close to just backing the favourite.
Not enough: none beats the bookmaker's margin, and nobody is in profit.
Return per £1 staked, settled to Starting Price. The three house entrants are shown alongside the chatbots: The Favourite (in AI Picks the Winner), plus Stablebet Edge and the Stablebet Model (both in AI Finds the Value Bets). Research, not tips.
Edge vs Model: one engine, two rules
Our engine produces one thing each morning: a winning chance for every runner on the card. AI Finds the Value Bets enters those numbers twice, because there are two honest ways to bet them, and they answer different questions.
Stablebet Edge
Backs the runner whose price sits furthest above the engine's view of its true chance: the biggest disagreement with the market on the card. That disagreement usually lives on unfancied horses, so expect long prices, long losing runs, and the question this rule exists to answer: are the model's boldest calls insight, or the places where it is most wrong?
The Stablebet Model
Backs the horse the engine rates most likely to win, which is often the favourite. It should win far more races than its stablemate, and can still lose money at Starting Price, because winning often and beating the bookmaker's margin are different things. Its question: is the specialist's best guess worth more than a chatbot's?
Both records build on this page from 17 July 2026, race by race. We will write up the comparison once there is a settled sample worth reading; until then the board above is the whole story.
Questions
What does AI Finds the Value Bets measure?
Whether an AI can use the market's view as an input and still find value. Each model sees the racecard and the crowd's implied chance for each runner (margin removed), then must back the horse it believes the market most underrates: its best value pick, with its own estimated chance logged alongside. Settled at Starting Price. Comparing it to AI Picks the Winner shows what the market's view does to a machine's judgement.
Why did the AI Finds the Value Bets record restart in July 2026?
Because our first version asked the wrong question. We showed the models the market's chances and asked which horse would win. With the answer on the card, every model quite rightly said the favourite, around 99% of the time. That was our error in framing, not theirs in reasoning, and it made the run worthless as a test. We reset the record on 17 July 2026 with the value question above; the old rows remain in the raw data for anyone who wants to audit them.
Why does your own engine get two entries here?
Because the same probabilities can be used two ways, and they answer different questions. Stablebet Edge backs the runner the engine rates furthest above the market's implied chance (its biggest edge, usually a longer-priced horse); the Stablebet Model backs the runner it rates most likely to win (often the favourite). Running both on the same races shows whether the engine's boldest disagreements with the market hold up better or worse than its straight win calls. Neither could honestly run blind: the engine is trained on market-derived features, so the market's view is baked into everything it produces.
Is AI Finds the Value Bets profitable?
The board is derived live and reports wins and losses honestly, and the record is young; it restarted in July 2026. Hunting value sounds like the road to profit, but the bookmaker's margin sits inside every price, and finding genuine value more often than you pay that margin is the hardest trick in betting. Watch the board rather than taking anyone's word for it. Nothing here is a tip.
Gamble responsibly.This page is research and entertainment, not betting advice. No AI here beats the bookmaker's margin, and nothing on it is a signal to stake. Betting should never be a way to make money. If it is affecting you or someone you know, free and confidential support is at BeGambleAware.org. 18+.